DENSITY OPERATOR AND APPLICATIONS IN NONLINEAR

AND QUANTUM OPTICS

Fam Le Kien

Department of Applied Physics and Chemistry,

University of Electro-Communications, Chofu, Tokyo 182-8585, Japan

Lecture notes for the fall semester of 2008

1
Contents

I. GENERAL THEORY OF THE DENSITY OPERATOR

1. Concept of a statistical mixture of states

2. Pure states. Introduction of the density operator
3. Statistical mixtures of states
4. Applications of the density operator

II. TWO-LEVEL ATOMS INTERACTING WITH A LIGHT FIELD

1. Atom+field interaction Hamiltonian

2. Optical Bloch equations
3. Absorption spectrum: saturation and power broadening
4. Field propagation equations
5. Susceptibility, refractive index, and absorption coefficient

2
 I. GENERAL THEORY OF THE DENSITY OPERATOR

1. Concept of a statistial mixture of states

Consider a quantum system. If we have complete information about the system, the
state of the system can be described by a state vector (wave function), denoted by |  .
Such a state is called a pure state.
It often happens that we don’t know the exact state vector of the system. In this case,
we say that we have incomplete information about the system. For example, a photon
emitted by a source of natural light can have any polarization state with equal probability.
Similarly, a system in thermal equilibrium at a temperature T has a probability
proportional to e En / kT of being in a state of energy En .
More generally, the incomplete information usually presents itself in the following
way: the state of this system may be either the state | 1  with a probability p1 or the
state | 2  with a probability p2 , etc… Obviously, we have

p1  p2  ...   pn  1. (1)
n

We then say that we have a statistical mixture of states | 1  , | 2  , ... with probabilities
p1 , p2 , … A statistical mixture of two or more than two different states is called a
mixed state. In general, a quantum state can be either a pure state or a mixed state.
We emphasize that a system described by a statistical mixture of states must not be
confused with a system whose state |  is a linear superposition of states:

|    Cn | n . (2)
n

For a linear superposition of states, there exist, in general, interference effects between
these component states. Such effects are due to the cross terms of the type CnCn' ,
obtained when the modulus of the wave function is squared. Meanwhile, for a statistical
mixture of states | n  , we can never obtain interference terms between these component
states.
We illustrate the above statements by considering a single particle in the coordinate
space R of the position vector r .
If the particle is in a linear superposition state  (r )   Cn n (r ), the probability of
n
finding the particle at a given point r is
P(r ) | (r) |2 |  Cn n (r) |2   CnCn*
' n (r ) n ' (r ).
*
(3)
n nn '

3
The above expression involves cross terms of the type CnCn' , which describe interference
between the component states  n (r ) .

However, if the particle is in a statistical mixture of states | n  with weight factors

pk , the probability of finding the particle at a given point r is

P(r )   pn | n (r ) |2 . (4)
n

The above expression shows no interference between the component states  n (r ) .

2. Pure states. Introduction of the density operator

We first examine the simple case where the state vector of the system is perfectly
known, that is, all the probabilities pn are zero, except one. The system is then said to be
in a pure state.

a) Description by a state vector (wave function)

i. State space
A pure quantum state of any physical system is characterized by a state vector,
belonging to a linear space S which is called the state space of the system.

ii. Ket vectors

An element, or vector, of the state space S is called a ket vector. It is represented by
the symbol    .

iii. Scalar product

With each pair of ket vectors    and   ' , taken in this order, we associate a
complex number, which is denoted by (    ,   ' ) and satisfies the various properties
(  ,   '  (  ',   * ,
(  , 1  1   2  2   1 (   ,  1   2 (   ,  2  (5)
(1  1   2  2 ,     1* ( 1 ,     2* ( 2 ,   
This complex number is called the scalar or inner product. The linear space with this type
of scalar products is called a Hilbert space.

iv. Bra vectors and dual space

Each ket vector    is associated with a bra vector, denoted by the symbol  | ,
which associates every ket vector   ' with a complex number which is denoted by
   ' and is equal to (  ,   ' . The set of bra vectors constitutes a vector space called
the dual space of the state space S.

4
 v. Basic properties of the ket and bra vectors
(1)  |  '   ' |   *.
(6)
(2) If |    1 | 1   2 | 2  then  | 1* 1 | 2* 2 | .

More generally, for the superposition state vector |    Cn | un  , we have

n

 |   un | Cn*. (7)
n

In the case of a single particle in the coordinate space R of the position vector r , the
ket vector    and the bra vector   are represented by the complex functions  (r ) and
 * (r ) , respectively, and the scalar product    ' is defined as    '    * (r) '(r)dr .
It is clear that    '   '    .

vi. Orthonormal basis

The state of a pure-state system can be described by a state vector |  . The set of
state vectors | un  is a basis set if an arbitrary state vector |  can be decomposed as

|    Cn | un . (8)
n

The set of basis states | un  is an orthonormal basis set if

un  um    nm . (9)

The coefficients Cn in Eq. (8) are called the probability amplitudes of the superposition
state |  in the basis {| un } . More exactly, we say that Cn is the probability amplitude
of the basis state | un  in the superposition state |  .

It follows from the expansion (8) and the orthonormality (9) of the basis states | un 
that the coefficients Cn of the state vector |  can be determined as
Cn  un |  (10)
We say that the state vector |  is normalized if  |   1 . If the state vector |  is
normalized, its coefficients satisfy the normalization condition

| C
n
n |2  1. (11)

From now we consider only normalized state vectors.

Note that the bra vector  | associated with the ket vector |  can be decomposed in
the form

5
  |   un | Cn*. (12)
n

With the use of the orthonormality of the basis states | un  , we find

Cn*   | un   un |   (13)

vii. Projection operators

We introduce the operator
P|un  | un un |, (14)

whose action on an arbitrary vector |V  is defined as

P|un  | V   (| un  un |) | V 
(15)
| un  ( un | V  | un  un | V 

As seen, the action of P|un  on |V  gives a vector parallel to | un  . In addition, we have

P|u2n   P|un  P|un 

 (| un  un |)(| un  un |)
| un  ( un | un  ) un | (16)
| un  un |
 P|u  .
n

The operator P|un  | un un | is called the projection operator onto the basis vector | un  .

viii. Completeness of the basis

The completeness of the set of the basis states | un  means that

P
n
|un   P|u1   P|u2   ...  1, (17)

that is,

| u u
n
n n | | u1  u1 |  | u2  u2 | ...  1. (18)

From now we use only the orthonormal basis. The properties of this basis are
summarized below:
 un  um    nm ,
 | u u
n
n n |  1. (19)

ix. Linear operators

A linear operator O associates each ket vector |   with a ket vector |  '  O |   and
satisfies the linearity

6
 O(1 | 1   2 | 2  )  1O | 1   2O | 2 . (20)
Similarly, the linear operator O associates each bra vector  | with a bra vector
 ' |  | O and satisfies the linearity
(1 1 | 2 2 |)O  1 1 | O  2 2 | O. (21)

For the superposition state vector |    Cn | un  , we have

n

 
O |   O   Cn | un     Cn O | un     CnO | un ,
 n  n n
(22)
 
 | O    Cn*  un |  O   Cn*   un | O    Cn*  un | O.
 n  n n

We consider the operator

O | A B |, (23)
whose action on ket and bra vectors is defined as
O |    | A B | |   | A   B |    | A B |  ,
(24)
 | O   | | A B |    | A   B |  | A B | .
We can show that
 | O |  '     | O  |  '   | A B |  '   | O |  '. (25)

x. Matrix form of a linear operator

With the help of the completeness relation (18), we can represent an arbitrary
operator O in the form
   
O    | un  un |  O   | um  um | 
 n   m 
  | un  un | (O | um  ) um |   | un  ( un | O ) | um  um | (26)
nm nm

   un | (O | um  ) | un  um |   ( un | O ) | um  | un  um |.
nm nm

Hence, we find
un | (O | um  )  (un | O) | um   un | O | um   Onm . (27)

The matrix Onm  un | O | um  is called the matrix form of the linear operator O . Every
operator O can be represented by a matrix Onm .

Every matrix Onm represents an operator O . Indeed, it follows from Eqs. (26) and
(27) that

7
 O   Onm | un um | . (28)
nm

We can generalize Eq. (27) to obtain

1 | (O | 2  )  (1 | O) | 2   1 | O | 2 . (29)

xi. Hermitian conjugate

Every operator O is associated with an operator O † , whose matrix is
(O† )nm  (Omn )*  (OT )*nm  un | O† | um   um | O | un *. (30)
Here the upper label T means the matrix transpose and the symbol * means the complex
conjugate. The operator O † is called the Hermitian conjugate of the operator O .
For O   Onm | un  um | , we have
nm

O †   Onm
†
| un  um |
nm

  Omn
*
| un  um | (31)
nm

  Onm
*
| um  un |.
nm

Note that O †   O .
†

We can show that

 | O  O† |, (32)

where O † | is the bra vector associated with the ket vector | O†   O† |   . Indeed,
since O   Onm | un  um | , we have
nm

 | O   | Onm | un um |  Onm  | un  um |. (33)

nm nm

On the other hand, since O †   Onm

*
| um  un | , we find
nm

 
| O †   O † |      Onm
*
| um  un |  |  
 nm 
  Onm | um  un |  
*
(34)
nm

  Onm
*
 un |   | um ,
nm

which yields
O † |  Onm  | un um |. (35)
nm

8
Comparison of Eq. (35) with Eq. (33) gives Eq. (32). Thus,  | O is a bra vector that is
associated with the ket vector O † |   . In the same way, we can say that the ket vector
O |   is associated with the bra vector  | O † .
Several useful properties of the Hermitian conjugates:
Property 1:
If  |  | O
(36)
then |    O † | .
Property 2:
If |    O | 
(37)
then  |  | O † .
Property 3:
 | O |  '*   ' | O† |  . (38)
Property 4:
If O  O1O2  On
(39)
then O †   *On†On†1  O1† ,
where  is a complex number and O1 , O2 , …, On are arbitrary operators.
Property 5:
If O  f ( A)
(40)
then O †  f ( * A† ),
where f is a real function, A is an arbitrary operator, and is a complex number.

xii. Hermitian operators

An operator O is called Hermitian if it is equal to its Hermitian conjugate O † , that is,
O†  O or *
Omn  Onm . (41)

xiii. Mean values of observables

Every observable is described by a Hermitian operator. Consider an observable
described by a Hermitian operator A . The matrix elements of the operator A in the basis
{| un } are given by

Anm  un | A | um   un |  A | um    (un | A) | um . (42)

Since A is a Hermitian operator, we have
Anm  Amn
*
. (43)

9
The mean value of A is
 A   | A | 
   
   Cn*  un |  A   Cm | um  
 n   m 
(44)
  CnCm  un | A | um 
*

nm

  Cn*Cm Anm .
nm

xiv. Schrödinger equation and Schrödinger picture of quantum mechanics

The time evolution of the state vector | (t ) of the quantum system is described by
the Schrödinger equation
d
i |  ( t )  H |  ( t )  , (45)
dt
where H is the Hamiltonian operator of the system. The Hamiltonian is the energy, an
observable. Therefore, it must be a Hermitian operator, that is, H †  H .
The solution to the Schrödinger equation (45) is given by
| (t )  exp(iHt / ) | (0). (46)
The corresponding bra vector is
 (t ) |  (0) | exp(iHt / ). (47)
The time-dependent mean value of the observable A is
 A(t )   (t ) | A | (t ). (48)

Consider the initial state vector | (0)   Cn (0) | un  . The probability amplitude
n

Cn (0) is given by Cn (0)  un | (0) For an arbitrary time t , we have

| (t )   Cn (t ) | un , (49)
n

where
Cn (t )  un | (t ) (50)
Inserting Eq. (49)into Eq. (45) gives
d
i  Cn (t ) | un    Cm (t ) H | um 
n dt m

  Cm (t ) | un  un | H | um  (51)
mn

  Cm (t ) un | H | um  | un .
mn

10
Hence, we obtain
d
i Cn (t )   H nmCm (t ). (52)
dt m

Here H nm  un | H | um  is the matrix of the Hamiltonian.

In terms of the probability amplitudes Cn (t ) , the mean value of the observable A at

the time t is given by
 A(t )   (t ) | A | (t )   Cn* (t )Cm (t ) Anm . (53)
nm

xv. Example of a two-level system

Figure 1: A two-level system.

The simplest quantum system is a two-level system, see Fig. 1. Such a system has two
stationary energy levels, denoted by Ea  a (upper level) and Eb  b (lower level).
The wave functions of these two levels are denoted by the state vectors | a (upper state)
and | b (lower state). These state vectors are normalized in modulus to unity and are
orthogonal to each other:
 a | a    b | b  1,
(54)
 a | b   b | a   0.
The completeness of the basis requires
| a a |  | bb | 1. (55)
In general, any state vector (wave function) of the two-level system can be written in the
form
|   Ca | a  Cb | b. (56)

Here Ca and Cb are the probability amplitudes of finding the system in states | a and
| b , respectively. They satisfy the normalization condition
| Ca |2  | Cb |2  1. (57)

11
The bra vector  | associated with the ket vector |  can be written in the form

 |  a | Ca*  b | Cb*. (58)

The mean value of an observable A is
 A   | A | 
   a | Ca*   b | Cb*  A  Ca | a   Cb | b 
(59)
| Ca |2  a | A | a  | Cb |2  b | A | b  Ca*Cb  a | A | b  Cb*Ca  b | A | a 
| Ca |2 Aaa  | Cb |2 Abb  Ca*Cb Aab  Cb*Ca Aba .
The total Hamiltonian is
H  H0  H I , (60)

where H 0 and H I are the free (unperturbed) and interaction parts, respectively. The
upper state | a and the lower state | b are the eigenstates of the free Hamiltonian H 0 ,
that is, H 0 | j  E j | j , where j  a, b . We can represent H 0 in the form

H 0  a | a a | b | bb | . (61)

Why does H 0 have the above form? The reason is the following: since
| j j | | aa |  | bb | 1 and H
j
0 | j  E j | j , we can expand H 0 as

H 0   i , j (| i i |) H 0 (| j  j |)
  i , j | i  j | (i | H 0 | j  )
(62)
  i , j | i  j | (i | j  E j )   i , j E j i , j | i  j |
  i Ei | i i |   i i | i i |.

We consider a particular case where the two-level system is free, i.e., H I  0 . In this
case, the Schrödinger equation is
d
i |   H 0 | 
dt (63)
 (a | a  a |  b | b b |) | .
It gives
d
i  Ca | a  Cb | b   (a | a a | b | bb |) Ca | a  Cb | b 
dt
 (64)
iC | a   iC | b   C | a    C | b.
a b a a b b

Hence, we find

12
 d
Ca  iaCa ,
dt
(65)
d
Cb  ibCb .
dt
The solutions to the above equations are
Ca (t )  e iatCa (0),
(66)
Cb (t )  e ibtCb (0).
Consequently, the explicit time-dependent expression for the state vector (wave function)
of the free two-level system is
| (t )  Ca (t ) | a  Cb (t ) | b  eiatCa (0) | a  eibtCb (0) | b. (67)
Note that the probabilities of finding the free two-level system in the upper and lower
levels are | Ca (t ) |2 | Ca (0) |2 and | Cb (t ) |2 | Cb (0) |2 , respectively. They are independent
of time. In the contrary, the interference described by the cross term
Ca (t )Cb* (t )  Ca (0)Cb* (0)ei (a b )t oscillates in time.
The mean value of an observable A at an arbitrary time t is
 A(t )   (t ) | A | (t )   Cn* (t )Cm (t ) Anm
nm

  C (0)Cm (0)e
*
n
i ( n m ) t
Anm
nm (68)
| Ca (0) |2 Aaa  | Cb (0) |2 Abb
Ca* (0)Cb (0)eiabt Aab  Cb* (0)Ca (0)e  iabt Aba .

Here ab  a  b is the transition frequency.

If the initial state is | (0) | a, then Ca (0)  1 and Cb (0)  0 . In this case, the state
vector at an arbitrary time is | (t )  eiat | a and the mean value of an observable A at
an arbitrary time is  A(t )  Aaa   a | A | a. The mean value is independent of time
because the initial state | a is an eigenstate of the system.
If the initial state is | (0) | b, then Cb (0)  1 and Ca (0)  0 . In this case, the state
vector at an arbitrary time is | (t )  eibt | b and the mean value of an observable A at
an arbitrary time is  A(t )  Abb  b | A | b. The mean value is independent of time
because the initial state | b is an eigenstate of the system.
If the initial state is a superposition state of | a and | b , with Ca (0)  0 and
Cb (0)  0, then  A(t ) in general varies in t.

13
 Exercises
Exercise 1: Consider a two-level system. Suppose that  is an operator such that
 | a | b and  | b  0 . Show that  can be written in the form  | b a |.
Exercise 2: By definition (O† )kl  Olk* , where O is an arbitrary operator and O † is its
Hermitian conjugate. Suppose O | un um | . Show that O† | um un | .
Exercise 3: Suppose O | A B | , where | A and | B are two arbitrary vectors. The
action of O | A B | on an arbitrary vector Q is defined as
O | Q  | A B | | Q | A   B | Q  | A B | Q . Show that O† | B A | .

Exercise 4: By definition O†  O T* . Suppose O  O1O2 . Show that O†  O2†O1† .

14
 Solutions
Exercise 1:
Since  | a | b and  | b  0 , we have
 | a a || b a |
and
 | bb | 0.
When we sum up the last two equations, we obtain
 (| aa |  | bb |) | ba |.
On the other hand, the completeness of the basis means that
| a a |  | bb | 1.
Hence, we find
 | b a |.
Using the orthonormality of | a and | b , namely  a | a   and  a | b   , we
can confirm that the operator  | b a | satisfies the relations  | a | b and
 | b  0 .

Exercise 2:
For O | un um | , we find

Okl   kn lm .
For X | um un | , we have

X kl   km ln .
We see that
X kl  Olk*  O†kl .

Hence, we obtain O†  X , that is,

O† | um un | .

Exercise 3:
For O | A B | , we have

Okl  ak bl* ,

15
where ak  uk | A and bl  ul | B .
For X | B A | , we have

X kl  bk al*.
We see that
X kl  Olk*  O†kl .

Hence, we obtain O†  X , that is,

O† | B A | .
Exercise 4:
For O  O1O2 , we have

O†  (O1O2 )†  (O1O2 )T *  (O2T O1T )*  O2T *O1T *  O2†O1† .

Generalization:
For O  O1O2  On , where  is a complex number and O1 , O2 , …, On are
arbitrary operators. Using the result of the above exercise, we can easily show that
O†   *On†On†1  O1† .

16
 b) Description by a density matrix
The relation
 A   | A |    Cn*Cm Anm (69)
nm

shows that the coefficients Cn enter into the mean values through quadratic expressions
of the type Cn*Cm . These quadratic expressions are simply the matrix elements of the
operator |  | , which is the projector onto the ket vector |  . Indeed, since

|    Cn | un ,
n
(70)
 |    un | Cn* ,
n

we have
Cn   un | 
(71)
Cn*   | un 
Hence, we find
Cn*Cm  um |  | un . (72)
It is therefore natural to introduce the density operator
 |  | . (73)
The density operator is represented in the {| un } basis by a matrix called the density
matrix whose elements are
 mn   um |  | un 
  um |  | un  (74)
 Cn*Cm .
The mean value of the observable A is then given by
 A   | A |    Cn*Cm Anm    mn Anm  Tr{ A}  Tr{A }. (75)
nm nm

Mathematically, the operator |  | is the projector onto the ket vector |  . Indeed,
with the use of the notation
P|  |  |, (76)

we see that the action of P|  on an arbitrary vector |V  gives a vector aligned along |  ,
namely,
P|  | V   (|  |) | V 
(77)
|  ( | V 

17
In addition, we have
P|2  P|  P| 
 (|  |)(|  |)
|  ( |  ) | (78)
|  |
 P|  .

The above properties indicate that P|  |  | is a projection operator. Since
P|2  P|  , we find that, in the case of pure states, we have  2   and hence,
Tr 2  Tr  1 [see Eq. (80)].
Note that the matrix  mn is a Hermitian matrix. Indeed, we have

 mn
*
 Cm* Cn   nm
 (79)
 †  .
The specification of  suffices to characterize the quantum state of the system. In
other words, the density operator  enables us to obtain all the physical predictions that
can be calculated from |  . To show this, we express in terms of the operator  the
conservation law of probability, the mean value of an observable, and the time evolution
of a quantum state.

i. Conservation law of probability

We find from the conservation law of probability that
Tr     nn   | Cn |2  1. (80)
n n

Thus, the sum of the diagonal elements of the density matrix is equal to 1.

ii. Mean value of an observable

For the mean value of an observable, we have the formula
 A   | A |    Cn*Cm Anm    mn Anm
nm nm (81)
 Tr{ A}  Tr{ A }.

iii. Schrödinger equation

We derive the time evolution equation for the density operator  . As known, the
Schrödinger equation is
d 1
|   H | . (82)
dt i

18
The Hermitian conjugate form of the above equation reads
d 1 1
 |   | H †    | H . (83)
dt i i
Hence, we have
d d
  |  |
dt dt
d  d 
  |    |  |    | 
 dt   dt 
1 1
 H |  |  |  | H (84)
i i 
1
 H  H 
i
1
 [ H ,  ].
i
Here
[O1, O2 ]  O1O2  O2O1 (85)

is the commutator between the operators O1 and O2 . Thus, the time evolution of the
density operator  is governed by the equation
d
i   [ H ,  ]. (86)
dt
This equation is called the generalized Schrödinger equation.

iv. Summary of the properties of the density operator of a pure state

 The properties of the density operator in the case of pure states are
 †   ,

Tr   1,
 A  Tr { A}  Tr { A }, (87)

i d   [ H ,  ].
 dt
The above properties are general, i.e., they are true also in the case of mixed states.
 In the case of pure states, there are two specific properties:

   ,
 2

 (88)
Tr   1.

2

These properties can be used as criteria to find out if a state is a pure state or not.

19
 v. Advantages of the description in terms of the density operator
A pure state can be described by a density operator as well as by a state vector (wave
function). Both descriptions are equivalent. However, the density operator presents a
number of advantages:
 First of all, two state vectors |  and ei |  describe the same quantum
state. In other words, there exists an arbitrary global phase factor for the state
vector. However, the state vectors |  and ei |  correspond to the same
density operator
 |  |
(89)
 ei |  | e i .
Thus, the use of the density operator eliminates the global phase.
 Second, the equation
 A  Tr { A}  Tr {A} (90)
is linear with respect to the density operator  .
However, the equation
 A   | A |  (91)
is quadratic with respect to |  .

vi. Example: the density operator of a pure state of a two-level system

We consider a two-level system in a pure state described by the state vector (wave
function)
|   Ca | a  Cb | b. (92)
The density operator of the system is
 |  | (Ca | a  Cb | b )(Ca*  a | Cb* b |). (93)
Taking the matrix elements, we get
 aa | Ca |2 ,
 ab  CaCb* ,
(94)
 ba   ab* ,
 bb | Cb |2 .
The matrix form of the density operator is
  ab   | Ca |2 CaCb* 
   aa  .
bb   CbCa* | Cb |2 
(95)
  ba
We see that  nm
*
  mn , that is,  †   . In addition, we have

20
 Tr  aa  bb | Ca |2  | Cb |2  1. (96)
The average value of an observable A is given by
 A  Tr{ A}  Tr{ A }
   mn Anm (97)
nm

  aa Aaa   bb Abb   ab Aba   ba Aab .

It is obvious that the diagonal matrix elements  aa and  bb are the probabilities of
being in the upper and lower states, respectively. The off-diagonal matrix elements  ab
and  ba , where  ab  ba
*
, describe the interference between the upper level | a and the
lower level | b in the state  . To illustrate more clearly the physical meaning of the off-
diagonal elements, we use the fact that the atomic polarization (average dipole moment)
of a single two-level atom is given by
P  e x  e | x | 
 e(| Ca |2 xaa  | Cb |2 xbb  Ca*Cb xab  Cb*Ca xba )
(98)
 e(Ca*Cb xab  Cb*Ca xba )
  ba d x   abd x* ,
where e is the electric charge of the electron and
d x  exab (99)

is the dipole matrix element. Here we have used the property xaa  xbb  0 , which is a
consequence of the fact that x is an odd function while | n ( x ) |2 (with n  a, b ) is, in the
case of atoms, an even function. The probability density | n ( x ) |2 is an even function
because so is the Coulomb potential in the atom. Thus, in the case of two-level atoms, the
off-diagonal elements  ab and  ba determine the atomic polarization.
As known, for a free two-level system, the Hamiltonian is
H 0  a | a a | b | bb | (100)
and the time evolution of the probability amplitudes is governed by the equations
d
Ca  iaCa ,
dt
(101)
d
Cb  ibCb .
dt
When we use the relation   C C* , we can easily show that the evolution equations
for the density-operator matrix elements of the free two-level system are

21
 d
 aa  0,
dt
d
 bb  0, (102)
dt
d
 ab  i (a  b )  ab .
dt
We can also derive the above equations directly from the generalized Schrödinger
equation (86). The solutions to Eqs. (102) are
  aa (t )  constant,

  bb (t )  constant, (103)
  i ( a b ) t
  ab (t )  e  ab (0).
As seen, when the two-level system is free, the diagonal matrix elements of the density
operator are constant. However, the off-diagonal matrix elements oscillate in time.

3. Statistical mixtures of states

a) Definition of the density operator
We now return to the general case where the system is in a mixture of states
| 1 ,| 2 ,...,| k ,... , with probabilities p1 , p2 ,..., pk ,... , where

0  pk  1,

 p  1.
 
(104)
k
k

The state vectors | k  of the components are normalized, that is,  k | k   1. However,
they are not necessarily orthogonal to each other.
The density operator of the statistical mixture of states is defined as
   pk ρk , (105)
k

where  k is the density operator corresponding to the state | k  , i.e.,

k | k  k | . (106)

b) Properties of the density operator

i. Hermitivity
Since the coefficients pk are real,  is obviously a Hermitian operator like each of
the  k :

22
  †   pk*  k†   pk  k   . (107)
k k

ii. Probability conservation law

The conservation law of probability gives
Tr    pk Tr  k
k
(108)
  pk  1.
k

iii. Mean values

The expression for the mean value of an observable is
 A   pk  A k
k

  pk Tr{A k }
k (109)
=Tr{ A pk  k }
k

=Tr{A }.
The averaging by means of the density operator, expressed by Eq. (109), has a twofold
nature. It comprises both the averaging due to the probabilistic nature of quantum
mechanics (even when the information about the object is complete) and the statistical
averaging necessitated by the incompleteness of our information. For a pure state, only
the first averaging remains. For a mixed state, both types of averaging are always present.

iv. Schrödinger equation for the density operator

We now derive an equation for the time evolution of the density operator. We assume
that the Hamiltonian H of the system is known. The state | k  satisfies the Schrödinger
equation
d
i | k   H | k . (110)
dt
Therefore,  k obeys the evolution equation
d
i  k  [ H ,  k ]. (111)
dt
Using the linearity of Eqs. (105) and (111), we find the following evolution equation for
the density operator  :

23
 d d  d 
i   i  pk  k   pk  i  k 
dt dt k k  dt 
  pk [ H ,  k ]  [ H ,  pk  k ] (112)
k k

 [ H ,  ].

v. Positivity
We see from the definition    pk ρk that, for any ket | u , we have
k

 u |  | u   pk  u |  k | u
k
(113)
  pk |  u | k  |2
k

and, consequently,
u |  | u  0. (114)
Equation (114) means that the density operator  is a positive operator.

vi. Summary of the properties of the density operator of a general quantum state
Thus, for pure states as well as statistical mixtures of states, we have
 †   ,

Tr   1,
 A  Tr { A}  Tr { A }, (115)

i d   [ H ,  ].
 dt

vii. Differences between the density operator of a mixed quantum state and that of a
pure state
The equations
 2  ,
(116)
Tr  2  1
are true for pure states but not true for mixed states. For mixed states, since  is no
longer a projector, we have
2   (117)

and, consequently, Tr  2  1. For a mixed state, according to comments (i) and (iii) at
the end of this section (see pages 26 and 27), we have
Tr  2  1. (118)
Thus, we have, in general, for pure and mixed states,

24
 Tr  2  1. (119)
The above equation can be used as a criterion to find out if a state is a pure state
(Tr  2  1) or a mixed state ( Tr  2  1 ).

c) Populations and coherences

What is the physical meaning of the matrix elements of the density operator  in the
basis {| un } ?

First, we consider the diagonal element  nn . We have

 nn   un |  | un 
  pk  un |  k | un 
k

  pk  un | k  k | un  (120)
k

  pk |  un | k  |2 .
k

We express the state | k  in terms of the basis {| un } as

| k    cn( k ) | un . (121)
n

This gives
cn( k )  un | k . (122)
Then Eq. (120) becomes
 nn   pk | cn( k ) |2 . (123)
k

It follows from Eq. (123) that  nn is a positive real number. The squared modulus | cn( k ) |2
is the probability of | un  in the pure state | k  . Therefore, Eq. (123) means that  nn is
the probability of | un  in the state  . For this reason, the diagonal matrix element  nn is
called the population of the state | un  . If the same measurement is carried out N times
under the same initial condition, where N is a large number, N  nn systems will be found
in the state | un  .

Now, we consider the non-diagonal element  nm . A calculation analogous to the

preceding one gives the following expression for  nm :

25
  nm   un |  | um 
  pk  un |  k | um 
k

  pk  un | k  k | um  (124)
k

  pk cn( k )cm( k )* .
k

The term cn( k )cm( k )* is a cross term. It expresses the interference effects between the states
| un  and | um  that can appear when the state | k  is a coherent linear superposition of
these states. According to Eq. (124),  nm is the average of these cross terms, taken over
all the possible states of the statistical mixture. If  nm is zero, this means that the
statistical average has canceled out any interference effects between | un  and | um  . On
the other hand, if  nm is different from zero, a certain coherence subsists between these
states. This is why the non-diagonal elements of  are often called coherences.

COMMENTS:

(i) The distinction between populations and coherences obviously depends on the
choice of the basis {| un } . Since  is Hermitian, it is always possible to find an
orthonormal basis {|  n } where  is diagonal. In this basis,  can be written as

   l | l  l | . (125)
l

The density operator  can thus be considered to describe a statistical mixture of the
orthonormal states |  n  with the probabilities  l . There are no coherences between the
states |  n  in this mixture. We have

Tr 2    l2
l
(126)
   l  1.
l

When one of the coefficients  l is equal to 1, all the other coefficients must be zero. In
this case, the state ρ is a pure state and Tr  2  1 . When all the coefficients  l are
smaller than 1, the state ρ is a mixed state and Tr  2  1 .
(ii) Assume that the Hamiltonian H is time-independent. If | un  are eigenvectors of
H, then
H | un   En | un ,
(127)
 un | H   un | En .
Hence, we obtain from Eq. (112)

26
 d
i  un   | um    un  H    H | um 
dt
  un  En    Em | um  (128)
  En  Em  un   | um 
that is,
d
i  nm  ( En  Em )  nm . (129)
dt
The solution to the above equation is
i
 ( En  Em ) t
 nm (t )  e 
 nm (0). (130)

In particular, we have nn (t )  constant .

Thus, the populations are constant, and the coherences oscillate at the Bohr
frequencies of the system.
(iii) We can easily prove that
nn mm | nm |2 . (131)
Indeed, according to Eqs. (123) and (124) and to the generalized Cauchy-Scharz-
Buniakowsky inequality, we have
  
 nn  mm    pk | cn( k ) |2    pk | cm( k ) |2 
 k  k 
2
 
   pk | cn( k )cm( k ) |  (132)
 k 
2

  pk cn( k )cm( k ) |  nm |2 .
k

It follows from Eq. (131) that  can have coherences only between states whose
populations are not zero.
It also follows from Eq. (131) that
Tr 2   |  nm |2
nm

   nn  mm
nm (133)
   nn   mm
n m

 1,
see Eq. (119).

27
 Exercises
Exercise 1: Consider a two-level system with density operator
1 3  14 0 
  | a a |  | b b |  3
.
4 4 0 4 
Show that the state of the system is a mixed state.

Exercise 2: Consider a two-level system with density operator

1 3 3 3  1 3

  | a a |  | b b |  | a b |  | b a |  4 4
 .
4 4 4 4  3
 
3
4 4

Show that the state  is a pure state.

Exercise 3: Consider two orthonormal vectors | a and | b . Consider a quantum

system (ensemble) Q1 that is prepared in the state | a with probability 1/4 and in the
state | b with probability 3/4.
Suppose we define
1 3
| 0  | a  | b
4 4
1 3
|1  | a  | b
4 4
Consider a quantum system (ensemble) Q2 that is prepared in the state | 0 with
probability 1/2 and in the state |1 with probability 1/2. Show that the quantum systems
Q1 and Q2 correspond to the same density operator.

Exercise 4: Consider two arbitrary vectors | A and | B Show that

Tr | A B |  B | A .

28
 Solutions
Exercise 1:
The square of the density operator of the system is
1 9 1 0
2  | a  a |  | b b |  16 9 
.
16 16 0 16 

We find Tr 2  1/ 16  9 / 16  10 / 16  5 / 8  1 . This inequality indicates that the state of

the system is a mixed state.

Exercise 2:
The square of the density operator of the system is
1 3 3 3
2  | a  a |  | b b |  | a  b |  | b a |
4 4 4 4
 14 3


 3 3 
4

 4 4 

 .
Hence we find Tr 2  Tr  1 . These equalities indicate that the state of the system is a
pure state.

Exercise 3:
The density operator of the system Q1 is
1 3
1  | a a |  | b b | .
4 4
The density operator of the system Q2 is
1 1
2  | 0 |  |1 | .
2 2
We can easily show that
1 1
2  | 0 |  |1 |
2 2
1 1 3  1 3  1 1 3  1 3 
  | a  | b    a |   b |   | a  | b    a |   b |
2 4 4  4 4  2 4 4  4 4 
1 3
 | a  a |  | b b |
4 4
 1.

29
Thus, two different realizations may correspond to the same statistical quantum
ensemble.
Since | a and | b are orthogonal to each other, they are the eigenstates of the density
operator   1   2 . In the contrary, | 0 and |1 are not orthogonal to each other.
Therefore, | 0 and |1 are not the eigenstates of the density operator  . In general, the
eigenvectors and eigenvalues of a density operator just indicate one of many possible
realizations that may give rise to a specific density operator.

Exercise 4:

With the help of the formula

| u u
n
n n |  1,

we find
Tr | A B |   un | A B | un 
n

   B | un  un | A
n

  B | A

30
 4. Applications of the density operator
a) System in thermal equilibrium
The first example we consider is borrowed from quantum statistical mechanics.
Consider a system in thermal equilibrium at the absolute temperature T. According to
quantum statistical mechanics, the density operator of the system is
  Z 1e H /kT . (134)
Here H is the Hamiltonian operator of the system, k is the Boltzmann constant, and Z is a
normalization constant chosen so as to make the trace of  equal to 1, that is,

Z  Tr {e H /kT }. (135)

We use the basis from the eigenvectors | un  of H. In this basis, we have

nn  Z 1 un | e H /kT | un   Z 1e E /kTn

(136)
and
nm  Z 1 un | e H /kT | um   0 for n  m. (137)
At thermal equilibrium, the populations of the stationary states are exponentially
decreasing functions of the energy, and the coherences between stationary states are zero.

b) Description of a part of a physical system

The most important application of the density operator is as a descriptive tool for
subsystems of a composite quantum system. Such a description is provided by the
reduced density operator. The reduced density operator is indispensable in the analysis of
composite quantum systems.
Suppose we have a composition of physical systems A and B. The state space of the
combined system AB is the direct product of the state spaces of the subsystems A and B.
By definition, the state space of AB contains the vectors of the type
| V AB  | V AV B  | V A  | V B  | V A  | V B  (138)

and also their superpositions, where | V A  and | V B  are arbitrary vectors in the state
spaces of A and B, respectively. The vector | V AB  has two parts, one belongs to A and
the other belongs to B. We have | V AB  | V BA  , | V AV B  | V BV A  ,
| V A  | V B  | V B  | V A  , and | V A  | V B  | V B  | V A  . The dimension of the state space
of AB is the sum of the dimensions of the state spaces of A and B.
The basis states of the combined system AB are
| u AB  | u Au B  | u A  | u B  | u A  | u B  (139)

where | u A  and | u B  are the basis states of the subsystems A and B, respectively.

31
 To label the individual basis states of the subsystems A and B, we use indices n A and
nB , respectively. Then, we can label the basis states of the combined system AB as

| unAB
AnB
 | unAA unBB  | unAA  | unBB  | unAA  | unBB  (140)

We describe the state of the composite system AB by a density operator  AB . The

matrix form of the density operator  AB is given by

 nABn
A B , m A mB
  unABAnB   AB | umABAmB 
  unAA unBB   AB | umAA umBB 
(141)
  
  unBB |  unAA   AB | umAA  | umBB  
  unBB |  unAA   AB | umAA  | umBB .

We assume that only part A is observed. Then information about part B is lost. This is
the so-called incomplete detection. Therefore, a statistical average over part B is
necessary. The state of part A is described by the reduced density operator
 A  TrB ( AB )
   unB |  AB | unB ,
(142)
n

where TrB is a map of operators known as the partial trace over system B. In the matrix
form, we have
 nA m    nABn
A A A B ,m AnB
. (143)
nB

Note that the partial trace of an operator of the type O AB | a1  a2 |  | b1 b2 | is given
by
TrB (| a1  a2 |  | b1  b2 |) | a1  a2 | TrB (| b1  b2 |)
(144)
| a1  a2 |  b2 | b1 .
Here | a1  and | a2  are any two vectors in the state space of A, and | b1  and | b2  are
any two vectors in the state space of B. In deriving the above expression, we have used
the formula Tr | 1 2 | 2 | 1  .
If the composite system is in a pure state, the incomplete detection process may cause
a part of the system to be in a statistical mixture. As an example, consider a two-level
atom initially prepared in the excited state. Assume that the atom interacts with a single-
mode radiation field that is initially prepared in the vacuum state.
The excited and ground states of the atom are indicated by | a and | b , respectively.
The n-photon state of the field is denoted by | n , where n  0,1,2,... . The basis states of
the composite system “atom+field” are | a;0 , | a;1 , | a;2 , … , | a; n , … , and | b;0 ,
| b;1 , | b;2 , … , | b; n , … .

32
 The initial state of the composite system is | a;0 . After a short time the atom has a
probability to make a transition to the ground state by spontaneous emission of a photon.
The state vector of the composite system at an arbitrary time t is then given by
| AB    | a;0   | b;1. (145)
The density operator of the composite atom+photon system is
 AB | AB  AB |
  | a;0   | b;1    a;0 |  *   b;1 |  * 
(146)
|  |2 | a;0 a;0 |  |  |2 | b;1 b;1 |
 * | a;0 b;1 |  * | b;1 a;0 | .
If we observe the state of the atom but not the emitted photon, then the atom will be
found in either the excited | a or the ground state | b ; however, it will no longer be in a
pure state. The new state can be described by the reduced density operator
 atom  Trph  AB  Trph | AB  AB |
(147)
|  |2 | a  a |  |  |2 | b b | .

Basic properties of the reduced density operator

Hermitivity: We can show that

  A †
  A,
or, (148)

 
*
A
nn '   nA' n .

Indeed, the matrix elements of the reduced density operator  A are given by

 nnA '   unA |  A | unA' 

  unA |   umB |  AB | umB  | unA' 
m

   u u |  AB | unA'umB 
A B (149)
n m
m

   nm
AB
,n ' m .
m

From the Hermitivity of  AB , we have   nm ,n ' m '    n ' m ',nm . Hence, we find
AB AB *

*
 
 
A *
    nm ,n ' m      nm ,n ' m 
AB AB *
nn '
 m  m (150)
   n ' m ,nm   n ' n .
AB A

33
Thus,  A is a Hermitian operator.

Normalization: We can show that

Tr A  1. (151)
Indeed,
Tr A    nnA    nm
AB
,nm
n n m (152)
 Tr AB
 1.
Thus, the probability conservation law is valid for the reduced density operator.
Average of an observable: Consider an observable O A that depends only on part A .
We can show that the average of this observable is given by the formula
O A   TrA AO A  TrAO A A. (153)
Indeed, by definition we have
O A   TrAB  ABO A
 TrATrB  ABO A
(154)
 TrA  TrB  AB  O A
 TrA  AO A  TrAO A  A .
Another way of proving is the following:
O A   TrAB  ABO A
   unm
AB
|  ABO A | unm
AB
 (155)
nm

   unAumB |  ABO A | unAumB .

nm

Since O A does not depend on part B , it does not change any vectors in the state space of
B , that is,
O A | unAumB   O A (| unA   umB 
 O A | unA   umB  (156)
 u O | u 
B
m
A A
n

Consequently, we have

34
 O A     unAumB |  ABO A | unAumB 
nm

   unA   umB |  AB | umB O A | unA 

nm

   unA   umB |  AB | umB }O A | unA  (157)

n m

 TrA{{TrB  AB }O A}
 TrA{ AO A}  TrA{O A  A}.

Thus, the average of O A can be determined from the reduced density operator  A by
using the same formula as for an arbitrary quantum system. This property makes the
reduced density operator  A useful for describing the state of part A.
Invalidity of the Schrödinger equation: In general, the Schrödinger equation is not
valid for the reduced density operator. An example is given below.

c) Spontaneous emission
Consider a two-level atom. If the atom is completely free, that is, if there is no
interaction and no spontaneous emission, the evolution of the density operator  of the
atom is governed by the generalized Schrödinger equation
d 1
  [ H 0 ,  ]. (158)
dt i
Here the Hamiltonian of the free atom is given by
H 0  a | a a | b | bb | . (159)
When spontaneous emission is included, the evolution of the density operator  of
the atom cannot be described by the Schrödinger equation. According to the Weisskopf-
Wigner theory, the reduced density operator  of the atom is governed by the master
equation
d 1 
  [ H 0 ,  ]       2         . (160)
dt i 2
Here  is the atomic decay rate, and   | ab | and   | b a | are the atomic
transition operators. The decay rate  is sometimes called the Einstein A-coefficient and
  
is given as   ab 3
| d |2 /3 0c3 , where d  e |  a | r | b | is the dipole moment of the
atom and ab  a  b is the atomic transition frequency. Equation (160) can be
represented in the form
  0  decay , (161)
where

35
 1
 0  [ H 0 ,  ],
i (162)

 decay        2         .
2
We find
 0  i a | a  a |    | a  a |  b | b b |    | b b |  ,
 (163)
 decay   | a a |   2 | b a |  | ab |   | a a | .
2
We can easily do some exercises to show that
 a |  0 | a   0,
 b |  0 | b  0, (164)
 a |  0 | b  iab  ab ,
and
 a |  decay | a    aa ,
 b |  decay | b   aa , (165)

 a |  decay | b    ab .
2
The equations of motion for the matrix elements of  can be obtained from Eq. (161):
d
 aa   aa ,
dt
d
 bb   aa , (166)
dt
d 
 ab  iab  ab   ab .
dt 2
It may be noted that  aa  bb  0 , that is, aa  bb  constant . This result is valid
only in the framework of the two-level atomic model. In the case where the atom has
three or more levels, the atom can decay from level a to levels other than b. In this case,
the sum of the populations of levels a and b is not conserved.
The solution of Eqs. (166) is found to be
 aa (t )   aa (0)e t ,
 bb (t )  1   aa (0)e t , (167)
 iabt t /2
 ab (t )   ab (0)e .

36
Thus, the upper-state population  aa and the coherence  ab decay exponentially with
the rates  and  / 2 , respectively. In other words, the decay rate of  ab is half the decay
rate of  aa .

37
 Exercises
Exercise 1: Consider an arbitrary operator O and two arbitrary vectors |V  and |U  .
Show that
Tr{O | V U |}  U | O | V .
Exercise 2: Consider an arbitrary operator O and a state vector |  . Show that
Tr{O |  |}  O.
Exercise 3: Consider an arbitrary operator O and two basis vectors | un  and | um  .
Show that
Tr{O | un um |}  Omn .
Exercise 4: Consider a density operator  and two basis vectors | un  and | um  .
Show that
Tr{ | un um |}  mn .

38
 Solutions
Exercise 1: Using the formula Tr{ | A B |}   B | A , we can show that
Tr{O | V U |}  U | O | V .
Exercise 2: Use the result of exercise 1 and the definition O   | O | .
Exercise 3: Use the result of exercise 1 and the definition Omn  um | O | un .

Exercise 4: Use the result of exercise 1 and the definition mn  um |  | un .

39
 3. TWO-LEVEL ATOMS INTERACTING WITH A LIGHT
FIELD
1. Atom+field interaction Hamiltonian

Figure 2: A two-level atom interacting with a light field.

We consider a two-level atom interacting with a light field, see Fig. 2. The total
Hamiltonian of the atom in the field is given by
H  H0  H I , (168)

where H 0 and H I are the free (unperturbed) and interaction parts.

The free part of the Hamiltonian is the operator for the energy of the atom when it is
free and is given by
H 0  a | a a | b | bb | . (169)

The projection operators  aa | a a | and  bb | bb | describe the populations

(probabilities) of levels a and b, respectively. Indeed, we have
 aa   Tr{ | a a |}   a |  | a  aa (170)
and
 bb   Tr{ | bb |}  b |  | b  bb . (171)

Here, the diagonal matrix elements  aa and  bb of the density operator describe the
populations of the atom in levels a and b, respectively. Note that the energy of the atom
when it is free is given by
 H 0   a | a  a |  b | b b |
 a  aa   b  bb  (172)
 a  aa  b  bb .

40
 We introduce the Pauli operator  z | a a |  | bb | , which describes the inversion of
the populations of the atom. Note that  z   aa  bb . In terms of the population
inversion operator  z , we have
1
| a  a | (1   z ),
2
(173)
1
| b b | (1   z ).
2
Hence, we find
ab 
H0   z  (a  b )
2 2
(174)
ab
  z  const,
2
where ab  a  b is the transition frequency of the atom. We can choose the middle
point between a and b as the origin for the zero energy. This leads to a  0 / 2 and
b  0 / 2 . Then, the constant is the above expression for the Hamiltonian H 0 is zero.
More general, a constant in the Hamiltonian can always be neglected because it
commutes with the density operator and the observables operators and consequently does
not affect their time evolution.
We now describe the atom-field interaction. The interaction part of the Hamiltonian is
the operator for the interaction energy between the atom and the field. We assume that
the field is linearly polarized along the x direction. We use the dipole approximation for
the interaction. Then, the interaction part of the Hamiltonian is given by
 
H I  d  E  exE
 e(| a  a | xaa  | b b | xbb  | a  b | xab  | b a | xba ) E
(175)
 e(| a  b | xab  | b a | xba ) E
 (| a  b |  | b a |)d x E.
Here we have used the properties xaa  xbb  0 , have introduced the notation d x  exab ,
which is called matrix element of the atomic dipole moment, and have assumed for
simplicity that xab and consequently d x are real parameters.
We present the electric component of the light field in the form

E  E0 cos t 
2
 e  e  E0.
1 it it
(176)

We introduce the Rabi frequency

  d x E0 / . (177)
Then, we obtain

41
 
H I   (| a  b |  | b a |)(e  it  eit )
2
(178)

  (     )(e  it  eit ).
2
Here we have introduced the upward transition operator
    ab | ab | (179)
and the downward transition operator
    ba | b a |. (180)
We note that, for a free atom, in the Heisenberg presentation, the transition operators
  | ab | and   | b a | oscillate as eiabt and eiabt , respectively. Indeed, for a free
atom we have     Tr   Tr | b a |  a |  | b  ab  ab (0)eiabt . Therefore, the
terms   eit | ab | eit and  eit | b a | eit quickly vary as ei (ab  ) t and ei (ab  ) t ,
respectively. Meanwhile,   eit | ab | eit and  eit | b a | eit slowly vary as
ei (ab  )t and ei (ab  )t , respectively.
Let’s come back to the interaction Hamiltonian (178). We neglect the fast rotating
terms | ab | eit and | b a | eit . This approximation is called the rotating-wave
approximation. The result is

HI   (  e  it    eit )
2
(181)

 (| a  b | e  it  | b a | eit ).
2
Appendix: Properties of the Pauli operators   ,   , and  z :
The operators   | ab | ,   | b a | , and  z | a a |  | bb | are called the Pauli
operators. The matrix forms of these operators are
0 1
   ,
0 0
 0 0
   , (182)
1 0
1 0 
z   .
 0 1 
The operators  x       and  y  i(     ) together with  z are the original Pauli
operators invented by W. Pauli to describe the spin 1/2 of an electron in the first half of
the last century.
We can easily show that

42
  2   2  0,
 z2  1,
1
    (1   z ),
2
1
    (1   z ),
2
 z     ,
 z     ,
and
  z    ,
  z    .
The above relations allow us to reduce any function f (  ,   ,  z ) of the Pauli operators
to a linear form of the type c   c   cz z .
In particular, we find the following commutation relations:
[ z ,   ]  2  ,
[ z ,   ]  2  ,
[  ,   ]   z .

Exercise
0
Consider a free two-level atom, with the Hamiltonian H  H 0   z . Use the
2
commutation relations for the Pauli operators to derive the Heisenberg equations for these
operators.

Solution
The Heisenberg equation for the observable O of a quantum system with the
Hamiltonian H reads
d i
O  [ H , O ].
dt 
When we use the commutation relations [ z ,   ]  2  , [ z ,   ]  2  , and

[  ,   ]   z and the Hamiltonian H  H 0  0  z , we find
2

43
 d i
 z  0 [ z ,  z ]  0,
dt 2
d i
   0 [ z ,   ]  i0  ,
dt 2
d i
   0 [ z ,   ]  i0  .
dt 2
The solutions to the above equations are  z (t )   z (0) ,   (t )    (0)exp(i0t ) , and
  (t )    (0)exp( i0t ) .

2. Optical Bloch equations

a) Evolution equations for the state vector of a pure state of a two-level atom
We first consider the case where the state of the atom is a pure state, described by a
state vector
|   Ca | a  Cb | b. (183)
The state vector |  satisfies the Schrödinger equation
d
i |   H |   ( H 0  H I ) | . (184)
dt
For the probability amplitudes Ca and Cb , we find the equations
d i
Ca  iaCa  e  itCb ,
dt 2
(185)
d i
Cb  ibCb  eitCa .
dt 2
The case of exact resonance:
Let consider the case of resonance, where the frequency ω of the field coincides with
the atomic transition frequency 0  a  b , that is,   0 . We change the fast
oscillating amplitudes Ca and Cb to the slowly varying amplitudes

ca  Ca eiat ,
(186)
cb  Cbeibt .
Then, Eqs. (185) yield
d i
ca  cb ,
dt 2
(187)
d i
cb  ca .
dt 2
The solutions for ca and cb can be written as

44
  t   t 
ca (t )  ca (0) cos    icb (0)sin   ,
 2   2 
(188)
 t   t 
cb (t )  cb (0) cos    ica (0)sin   .
 2   2 
This solution is general with respect to the initial condition.
Example 1:
We consider the situation where the atom is initially in the ground state b, that is,
ca (0)  0 and cb (0)  1 . In this case, we find

 t 
ca (t )  i sin   ,
 2 
(189)
 t 
cb (t )  cos   .
 2 
Hence, we have
 t 
Ca (t )  i sin   e  iat ,
 2 
(190)
 t 
Cb (t )  cos   e  ibt .
 2 
The populations of the levels, described by the diagonal matrix elements  aa and  bb ,
are found to be
t 
 aa (t ) | Ca (t ) |2  sin 2  ,
 2 
(191)
 t 
 bb (t ) | Cb (t ) |  cos   .
2 2

 2 
As seen, the populations oscillate in time, with the frequency Ω. Such oscillations are
caused by the driving field. They are different from the optical oscillations, which occur
with the frequency  . The oscillations of the populations, caused by the driving field, are
called the Rabi oscillations, and Ω is called the Rabi frequency. The Rabi oscillations
mean that the atom jumps forth and back between the ground and excited states.
Meanwhile, the coherence (interference) is described by the off-diagonal matrix
element
 ab (t )  Ca (t )Cb* (t )
 t   t 
 i sin   cos   e  i (a b ) t (192)
 2   2 
i i
 sin(t )e  i0t  sin(t )e  it .
2 2

45
Note that the coherence  ab oscillates with the optical frequency 0   , and the slowly
varying envelope of the coherence oscillates with the Rabi frequency Ω. We usually have
   . Thus, the Rabi oscillations are usually very slow compared to the optical
oscillations.
Example 2:
We consider the situation where the atom is initially in the excited state a, that is,
ca (0)  1 and cb (0)  0 . In this case, we find

 t 
ca (t )  cos  
 2 
(193)
 t 
cb (t )  i sin   .
 2 
Hence, we have
 t 
Ca (t )  cos   e  iat ,
 2 
(194)
 t 
Cb (t )  i sin   e  ibt .
 2 
The populations of the levels, described by the diagonal matrix elements  aa and  bb ,
are found to be
t 
 aa (t ) | Ca (t ) |2  cos2  ,
 2 
(195)
 t 
 bb (t ) | Cb (t ) |  sin   .
2 2

 2 
As seen, the populations oscillate in time, with the Rabi frequency Ω. Meanwhile, the
coherence (interference) is
 ab (t )  Ca (t )Cb* (t )
 t   t 
 i sin   cos   e  i (a b ) t (196)
 2   2 
i i
  sin(t )e  i0t   sin(t )e  it .
2 2
We also observe that the coherence  ab oscillates with the optical frequency 0   , and
the slowly varying envelope of the coherence oscillates with the Rabi frequency Ω.
Example 3:
We consider the situation where the atom is initially in a linear superposition of the
excited state a and the ground state b, with the probability amplitudes ca (0)  1 / 2 and
cb (0)  1 / 2 . In this case, we obtain the solution

46
 1   t   t  
ca (t )  cb (t )   cos    i sin    , (197)
2  2   2 
which leads to
1   t   t    iat
Ca ( t )  cos  2   i sin  2   e ,
2     
(198)
1   t   t  
Cb (t )   cos    i sin    e  ibt .
2  2   2 
Hence, we find that the populations of the levels are constant in time:
aa (t )  bb (t )  1/ 2. (199)
We also find that the coherence is oscillating in time, with a constant envelope:
1 1 1
 ab (t )  Ca (t )Cb* (t )  e i (  )t  e i t  e it .
a b 0
(200)
2 2 2

b) Evolution equations for the density operator of a general quantum state of

a two-level atom with no decay
We now consider the case where the state of the atom is an arbitrary state, described
by a density operator ρ. The density operator  is governed by the Schrödinger equation
d 1
  [ H ,  ]. (201)
dt i
To obtain the evolution equations for the matrix elements of  , we can use Eq. (201)
directly. For this purpose, we write
d
  i (O  O † ), (202)
dt
where
1
O H . (203)

In the matrix form, we have
d
  i (O  O † )  i (O  O
*
). (204)
dt
Let’s calculate the matrix elements O . When we use the explicit expression for the
Hamiltonian H  H 0  H I , we find

47
 1 1
O H   ( H 0  H I )
 
(205)

 a | a  a |   b | b b |   (| a  b |  e it  | b a |  eit ).
2
We can easily show that

Oaa  a  aa   ba e  it ,
2

Obb  b  bb   abeit ,
2
(206)

Oab  a  ab   bbe  it ,
2

Oba  b  ba   aa eit .
2
When we insert the above matrix elements into Eq. (204), we obtain
d i
 aa  (e  it  ba  eit  ab ),
dt 2
d i
 bb   (e  it  ba  eit  ab ), (207)
dt 2
d i
 ab  iab  ab  e  it (  aa   bb ).
dt 2
Note that we can also derive Eqs. (207) with the use of Eqs. (185) and the definitions
 aa | Ca |2 ,
 bb | Cb |2 ,
(208)
 ab  CaCb* ,
 ba   ab* .
We introduce the notation     ab for the detuning of the field. It is convenient to
use new variables
 ab  eit  ab ,
 ba  e  it  ba ,
(209)
 aa   aa ,
 bb   bb .
In terms of these variables, we have

48
 d i
 aa  (  ba   ab ),
dt 2
d i
 bb   (  ba   ab ), (210)
dt 2
d i
 ab  i ab  (  aa   bb ).
dt 2
The above equations are called the optical Bloch equations for two-level atoms without
decay.

Example:
We consider the situation where the atom is initially in a statistical mixture of the
excited state a and the ground state b, with the weight factors pa  pb  1 / 2 . In this case,
the matrix elements of the initial density of the atom are given by
 aa (0)   bb (0)  1 / 2,
(211)
 ab (0)   ba (0)  0.
Consequently, we have
 aa (0)   bb (0)  1 / 2,
(212)
 ab (0)   ba (0)  0.
The Bloch equations show that the above values are true for any time t, that is,
 aa (t )   bb (t )  1 / 2,
(213)
 ab (t )   ba (t )  0.
Note that the above solutions and the solutions for example 3 in the previous subsection
give the same populations but different coherences.
Exercise
Derive Eqs. (207) from Eqs. (185). Such a derivation is valid only for the case of a
pure state. However, the result is general: it coincides with that for an arbitrary state.
Solution
Equations (185) and their complex conjugates can be written as

49
 d i  it
Ca  iaCa  e Cb ,
dt 2
d * i
Ca  iaCa*  eitCb* ,
dt 2
d i
Cb  ibCb  eitCa ,
dt 2
d * i 
Cb  ibCb*  e  itCa* .
dt 2
When we use the above equations and the relations
 aa | Ca |2 ,
 bb | Cb |2 ,
 ab  CaCb* ,
 ba   ab* ,
we can easily derive Eqs. (207).

c) Evolution equations for the density operator of a system with decay

We now include the spontaneous emission of the atom into our treatment. In this case,
the state of the atom is described by a reduced density operator  , which is governed by
the master equation
d 1
  [ H ,  ]   , (214)
dt i
where the last term describes the decay and is given by

   [     2        ]. (215)
2
We can rewrite Eq. (214) as
  (  )coherent  (  )decay , (216)
where
1
(  )coherent  [H ,  ] (217)
i
and

(  )decay     [     2        ]. (218)
2
In the matrix form, we have
  (  )coherent  (  )decay . (219)

50
The matrix elements (  )coherent are the matrix elements of the operator
1
[ H ,  ] . They are given by Eqs. (207). The matrix elements (  )decay are
(  )coherent 
i

the matrix elements of the operator (  )decay     [     2        ].
2
They are given by the expressions on the right-hand side of Eqs. (166). Indeed, we can
easily show that
  aa decay  (  )aa   aa ,
  bb decay  (  )bb   aa , (220)

  ab decay  (  )ab    ab .
2
To obtain the evolution equations for the matrix elements of  , we add to Eqs. (207) the
decay terms (220). This procedure leads to
d i
 aa   aa  (e  it  ba  eit  ab ),
dt 2
d i
 bb   aa  (e  it  ba  eit  ab ), (221)
dt 2
d   i
 ab    iab    ab  e  it (  aa   bb ).
dt  2 2
We introduce the notation     ab for the detuning of the field. It is convenient to use
new variables
 ab  eit  ab ,
 ba  e  it  ba ,
(222)
 aa   aa ,
 bb   bb .
Using these variables, we obtain
d i
 aa   aa  (  ba   ab ),
dt 2
d i
 bb   aa  (  ba   ab ), (223)
dt 2
d   i
 ab   i    ab  (  aa   bb ).
dt  2 2
The above equations are called the optical Bloch equations for two-level atoms with
decay.
The above optical Bloch equations play an important role in optics. In particular, they
are used to calculate the absorption spectrum of the atoms.

51
 3. Absorption spectrum: saturation and power broadening
We introduce the population difference
w   aa  bb (224)
and the optical coherence
c   ab . (225)
Then, we can rewrite Eqs. (223) as
d
w  ( w  1)  i(  c*   c ),
dt
(226)
d   i
 c   i    c  w.
dt  2 2
We consider the adiabatic (steady-state) regime where
dw d  c
  0. (227)
dt dt
In this regime, Eqs. (226) yield
w  i(  c*   c )  ,
  i (228)
 i    c  w.
 2 2
The solutions to the above equations are
1
w (229)
1 s
and
i
c  . (230)
2(  / 2  i )(1  s)
Here s is the saturation parameter and is given by
2 2 / 2 s0
s  2  , (231)
2 |  / 2  i |    / 4 1  (2 / ) 2
2 2

where s0 is the on-resonance saturation parameter and is defined as

2 2
s0  . (232)
2

52
For low saturation, s  1, the population is mostly in the ground state ( w  1 ). For
high saturation, s  1, the population is almost equally distributed between the ground
and excited states ( w  0 , i.e. aa , bb  1 / 2 ).

The parameter s0 can be written in another form

I
s0  . (233)
Is
Here
I  c 0 E02 / 2 (234)
is the intensity of the laser beam and
c 0  2  2
Is  (235)
2 2d x2
is the so-called saturation intensity. According to the Weisskopf-Wigner theory, we have

  03 | d |2 /3 0c3  03d x2 /  0c3. (236)

Here we have introduced for convenience a new notation 0  ab  a  b . With the
help of Eq. (236), we can show that
I s   hc / 03   hc / 03 , (237)
where
  1/  (238)
is the lifetime of the excited state and 0 is the atomic resonant wavelength.
The population of the excited state is given by
1 s s0 / 2
 aa  (1  w)   . (239)
2 2(1  s) 1  s0  (2 / ) 2
Since the population in the excited state decays at a rate  , the total scattering rate  scatt
of light from the laser field is given by
s0 / 2
 scatt   aa  . (240)
1  s0  (2 / )2
Note that, at high intensity,  scatt saturates to  / 2 .
Equation (240) can be rewritten as
s0 /2
 scatt  , (241)
1  s0 1  (2 /  ')2
where

53
  '   1  s0 . (242)

The dependence of the scattering rate  scatt on the detuning  is shown in Fig. 3 for
several values of the saturation parameter s0 . This dependence describes the absorption
spectrum. The width of the spectral profile is characterized by  ' . Note that the width
 ' increases with increasing intensity of the field. This phenomenon is called the power
broadening of the spectral profile.
The power broadening is a direct result of the fact that, for large s0 , the absorption
continues to increase with increasing intensity in the wings, whereas, in the center, half of
the atoms are already in the excited state. The absorption in the center is saturated,
whereas in the wings it is not.

Figure 3: Scattering rate  scatt as a function of the detuning for several values of the saturation
parameter s0 .

The scattering results in intensity loss when the beam travels through a sample of
atoms. The amount of scattered power per unit of volume is n scatt , where n is the
number density of the atoms. Thus, we have
dI s0  / 2 n(  / 2)( I / I s )
 n scatt  n 
dz 1  s0  (2 / ) 2
1  s0  (2 / ) 2 (243)
 n I   I .

54
Here  is the scattering cross section and   n is the absorption coefficient. These
coefficients are given by
  1
  . (244)
n 2 I s 1  s0  (2 / )2

c 0  2 2  hc
When we use the expressions I s   , we can rewrite Eq. (244) as
2 2d x2 03

  2d x2  1
  
n   0c  1  s0  (2 / ) 2
(245)
2 1
 0 .
2 1  s0  (2 / ) 2
For low intensity, s0  1 , we have

 2d x2  1
  
n   0c  1  (2 / ) 2
(246)
2 1
 0 .
2 1  (2 / ) 2
In this regime, the absorption coefficient  is independent of the field intensity I.
Therefore, the solution for the field intensity is
I ( z )  I 0 e  z . (247)

At exact resonance (   0 ), the cross section (89) reduces to


02
  . (248)
n 2
In the case where I  I s , the absorption coefficient  tends to zero. This does not
mean the vanishing of the absorption. Indeed, in the limit
I  I s (249)
we have
 I  n / 2. (250)
Hence, Eq. (243) yields
dI
  I  n(  / 2). (251)
dz
The solution to the above equation is
I  I (0)  n( / 2) z. (252)

55
According to the above equation, the field intensity will linearly reduce with increasing
propagation length. However, after the field intensity I becomes small as compared to
the saturation intensity I s , the field intensity I will exponentially decrease with
increasing propagation length z.

4. Field propagation equation

We study the propagation of the field along the z direction. The one-dimensional
wave equation reads
 2 2  2 1 2

 z 2 c 2t 2  E   P  P, (253)
 
0
t 2  0c 2 t 2
where P is the polarization density. We expand the field and the polarization in the forms
1
E  E0e  i (t kz )  E0*ei (t kz )  ,
2
(254)
1
P   P0e  i (t kz )  P0*ei (t kz )  .
2
Here  and k   / c are the frequency and wave number of the field, respectively.

We introduce new variables

  z,
(255)
  t  z / c.
In terms of these variables, we have

E
1
2
 E0e  i  E0*ei  ,
(256)
P   P0e  i  P0*ei  .
1
2
We also have
  
  ,
z  c
(257)
 
 ,
t 
and, consequently,
2 2 2 2 2
   ,
z 2  2 c  c 2 2
(258)
2 2
 .
t 2  2
On substituting Eqs. (256) and (258) into Eq. (253), we find

56
  2 2 2 2i  
e  i  2   E0  c.c.
  c  c  
(259)
 i 1  2  
e 2 
 2i   2  P0  c.c.
 0c   2
 
The above equation will be satisfied if
 2 2 2 2i   1  2  2
  2  c   c   0  c 2   2  2i     P0 .
E  (260)
  0  
We assume that the envelopes E0 and P0 vary slowly in space and time. When we
keep only the lowest nonvanishing order on each side of Eq. (260), we obtain
 i
E0  P0 . (261)
 2 0c
Since   z , we can rewrite Eq. (261) as
 i ik
E0  P0  P0 , (262)
z 2 0c 2 0
keeping in mind that the partial derivative with respect to z in Eq. (262) is taken under the
condition   t  z / c  constant . Equation (262) is called the propagation equation for
the field in the slowly varying envelope approximation.

5. Susceptibility, refractive index, and absorption coefficient

We have
P0   0  E0, (263)

where  is the susceptibility. We write

   ' i  '', (264)
where  ' and  '' are the real and imaginary parts of  . Then, Eq. (262) becomes
 i 
E0  (  ' i  '') E0  (i  '  '') E0 . (265)
z 2c 2c
The real part  ' determines the shift of the wave number of the field, while the
imaginary part  '' determines absorption. The refractive index of the medium is defined
by
'
nref  1  , (266)
2
while the absorption coefficient is defined by

57
 
  ''  k  ''. (267)
c
In terms of nref and  , we can rewrite Eq. (265) as

  
E0  ik (nref  1)   E0 . (268)
z  2
At low intensity, we can ignore the dependences of nref and  on the field intensity. In
this case, the solution of Eq. (268) for the field envelope E0 is

E0  Aeik ( nref 1) z e z /2 . (269)

When we use the presentation A | A | ei and the formula,

1
E  E0e i (t kz )  E0*ei (t kz )  , (270)
2
we obtain
E | A | e z /2 cos(t  knref z   ). (271)

The expression (271) explains why nref is called the refractive index and why  is called
the absorption coefficient.

We now calculate  , nref , and  . As shown in Sec. I.2.e, the polarization of a single
two-level atom is P  d x ( ab  ba ). Here, we have assumed for simplicity that d x is a
real parameter. For a medium with the atomic density n, the polarization density of the
medium is given by
P  nd x ( ab  ba )  nd x ( cei  c*ei ). (272)
Consequently, the envelope of the polarization density is
P0  2nd x c . (273)
Inserting Eq. (230) into Eq. (273) yields
i
P0  2nd x
(   2i )(1  s )
(274)
2nd x2 E0
i .
 (   2i )(1  s )
Hence, we have
P0   0  E0, (275)
where the susceptibility is given by

58
 P0

 0 E0
2nd x2 1
i (276)
 0 (   2i )(1  s )
2nd x2   2i
i .
 0 1  s0  (2 / ) 2
2

Then, we find
 4nd x2  
 '   2 
(277)
  0  1  s0  (2 / )
2

and
 2nd x2  1
 ''    . (278)
 
 0  1  s0  (2 /  ) 2

With the help of Eqs. (266) and (267), we obtain

 2nd x2  
nref  1   2 
  0  1  s0  (2 / )
2

(279)
n 3  /
 1  02
4 1  s0  (2 / ) 2
and
 2nd x2  1
  
  0c  1  s0  (2 / )
2

(280)
n 2 1
 0 .
2 1  s0  (2 / ) 2

Here we have used the expression   03d x2 /  0c3 and the approximation   0 .
Equation (280) is in agreement with Eq. (245). Note that nref and  depend on the field
intensity via the saturation parameter s0 .

59
Figure 4: Real and imaginary parts of the susceptibility as functions of the normalized detuning
 /  of the field for the saturation parameter s0  0.1 .

According to Eqs. (266) and (267), the real part  ' and the imaginary part  '' of the
susceptibility determine the refractive index nref and the absorption coefficient  . We
illustrate in Fig. 4 the frequency dependence of  ' and  '' . The figure shows that  ' is
large in the region  /   1. However, in this region,  '' is also large. Thus, in a
medium of two-level atoms, a high refractive index is accompanied by large dispersion.

60